Registration And Fusion Of Preoperative And Postoperative CT Volume Data Of Pedicle Screw Implantation

Pedicle screw implantation surgery requires careful insertion of screws into the spine to avoid complications.The surgical navigation system guides the implementation during the operation through medical images,which improves the accuracy of screw implantation.However,the current surgical navigation system is not completely safe,so it will be of great use to evaluate the postoperative images.Since postoperative CT images have serious metal artifacts and the preoperative CT images have clear bone tissue,the registration and fusion of the preoperative and postoperative CT images can use the clear bone tissue information of the preoperative images to make up for the information lost in the postoperative images due to metal artifacts,so as to provide the doctor with more judgment basis.The precise registration and the fusion after registration of preoperative and postoperative CT images are studied in this thesis.The implantation of screws causes the deformation of the spine,and the rigid registration based on the entire spine cannot obtain an accurate registration result,so it is necessary to register a single vertebra.This thesis studies the vertebral segmentation of CT spine images,preoperative and postoperative CBCT image registration,and CBCT image fusion,and finally realizes a complete process of preoperative and postoperative CT image registration and fusion.The main work of this thesis is as follows:First,research the spinal CT image segmentation for vertebral separation: in view of the difficulty of segmentation caused by the blurred boundary of the CT spine image,this thesis designed a deep convolutional neural network based on the U-Net network architecture to segment the vertebrae of CT and CBCT images.Experimental results show that the network can efficiently segment the vertebrae in CT and CBCT images,and there is no adhesion between adjacent vertebrae in the segmentation results,which can be used for subsequent registration research.Second,the research of preoperative and postoperative CBCT image registration: In the registration methodology of this thesis,each vertebra in CT was treated as a sub-volume and transformed individually.Research the similarity measurement function that can overcome the interference of metal objects,and the search strategy suitable for global registration,and solve the possible mismatch problems.A segmented registration experiment was performed to compare the registration performance of the three similarity measure functions,and verify the search performance of the optimization method combining the simulated annealing method and the Powell algorithm.The experimental results show that the normalized mutual information can achieve accurate registration of the images in this thesis,and the search strategy adopted in this thesis can converge to a position close to the optimal solution.The segmented registration algorithm in this thesis does not reduce the registration accuracy when the initial position is far from the target position.The translation error of the registration is less than 1mm,and the rotation error is less than 2°.An overall constraint solution was designed to solve the vertebra mismatch problem,and experiments were performed to compare the performance of the algorithm in this thesis and the registration of the entire spine.The registration accuracy of the algorithm in this thesis is much higher than that of the registration method of the entire spine.Third,fusion of the registered preoperative and postoperative CBCT image: In order to synthesize the information of preoperative and postoperative images,the fusion performance of the Laplacian pyramid fusion method was studied.The research found that for the images in this thesis,when the Laplacian pyramid is decomposed to the seventh layer,the two images achieve the best fusion result.The false-color display method was used to display the steel nails in the fusion result,and the resulting image clearly reflected the positional relationship between the metal object and the bone tissue.The algorithm in this thesis performs well in preoperative and postoperative image fusion.